Pinned Tweet
This is the real me, but the background is courtesy of Gemini 4 Argon.
I showed Argon the two green-screen stills below and asked it to 1) build me a room in Blender, 2) recreate the camera angles, 3) use Nano Banana to 'skin' the Blender background to match the foreground, and 4) key it back into the original video.
The result: an accurate background across both angles.
Meet Gemini 4 Neon — our newest model that brings frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows.
Neon brings frontier-level performance to your everyday work and is built to handle complex coding and problem-solving without the wait.
How, you might ask? ⬇️
— Smarter coding: Multi-file debugging and refactoring with targeted, surgical edits
— Grounded answers: Excels at multi-step instruction following and builds on our ongoing progress to curb fabrication, recording our lowest fabrication rate yet in internal Google Antigravity testing
— 1M-token context window: Synthesize massive projects, tables, and charts at once
— Adjustable thinking: Balance and adjust speed and depth on the fly with 4 thinking levels
Watch Gemini 4 Neon run a 12-hour /swarm in @antigravity to reconstruct Alan Turing’s 1936 study from archival photos — complete with a running Universal Turing Machine, the blueprint for modern computing.
George Kenwright retweeted
Meet Gemini 4 Neon :) Neon made this zero-library vanilla JavaScript 3D animation bringing John Keats' 1819 poem "To Autumn" to life in 12fps painterly impasto brushwork 🖼️ the most beautiful thing I have seen this week.
incredible work by @GeokenAI 🔥
George Kenwright retweeted
Gemini 4 Neon is rolling out now.
Sitting between Carbon and Hydrogen, it brings frontier intelligence and fast task completion for complex software engineering, knowledge work, and science.
I asked Gemini 4 Carbon to crack an unread 1466 Sforza Renaissance cipher. Unaided, it used its vision and research capabilities to decipher the text.
Gemini:
- transcribed 15th-century cipher characters from manuscript scans
- ran blind statistical decoding attempts
- pivoted to searching digitized European archives for clues
- found a handwritten flyleaf chart in Paris showing how Milan’s cipher office built syllable tables
- widened its search to all Sforza keys and spotted an unindexed gap in Vienna Codex 2398
- read Tranchedino’s 15th-century Latin index in Vienna and found Giovanni Bianco’s 1465 key on folio 47r
- mapped the key signs to decode all 18 encrypted lines in Italian
- uncovered warnings of Savoyard reprisals after an ambush on the new Duke of Milan, plus Louis XI’s secret three-marriage plan
Full details:
fofr.ai/bianco-1466-cipher
George Kenwright retweeted
Gemini 4 Carbon, Hydrogen, and the Decisions API in one game: Parking Bench.
Carbon built the tabletop puzzle in @antigravity and you race 1v1 against Hydrogen or Gemini Decision 1 on the Gemini API across 5 levels.
Both models play pure vision off a single board PNG with zero grid state.
George Kenwright retweeted
as an architecture nerd I enjoyed making this one
Replying to @Google
Public planning documents are usually a maze of floorplans and tables.
@cobley_ben used Carbon’s new /swarm skill in @Antigravity to explore them instead, turning PDFs into an interactive 3D recreation of our Platform 37 office in London in Three.js 🌆
George Kenwright retweeted
Gemini 4 Carbon is out today - spent a bit of time testing its long-horizon agentic coding on a personal experiment: rebuilding ~220m of downtown Los Gatos in Blender (~50 storefronts + 3D interiors), w/ ~zero manual modeling.
Quick look under the hood 🧵 (1/5)
Gemini 4 Carbon is great at agentic research and 3D modelling. So what happens when you combine them? @cobley_ben and I have been using agents to research and rebuild 2000 years of London history.
Gemini 4 Carbon made an Evolution of London simulation, built by @GeokenAI and myself
This version uses OpenHistoricalMap for the old bits and OSM Buildings for the new bits, Carbon fills in the gaps
George Kenwright retweeted
Gemini 4 Carbon made an Evolution of London simulation, built by @GeokenAI and myself
This version uses OpenHistoricalMap for the old bits and OSM Buildings for the new bits, Carbon fills in the gaps
Carbon:
- wrote a custom CUDA/C++ path tracer from scratch
- integrated MuJoCo rigid-body physics
- designed and tuned a 60-second Rube Goldberg machine (120,000 gravity and contact steps in MuJoCo, no keyframes)
- made the 3D studio and synced a continuous camera take to a custom Lyria 3.5 track
George Kenwright retweeted
I took @simonw's benchmark “pelican riding a bicycle” way too far using Gemini 4 Carbon and a /swarm in @antigravity to build a playable game.
Multi-agent critique loops ran across 5 waves to get this pelican biking across our Mountain View campus.
What's most interesting is the approach they took.
The subagents generated Nano Banana and Omni references to rig the assets in headless Blender before wiring the Three.js bike physics, creating custom WebGL2 shaders, and finally produced procedural Web Audio SFX with a Lyria track.
Try it out on the playable link in the reply 👇
George Kenwright retweeted
3D Pagoda
www.teapawt.org/antigravity/status/211…
Replying to @antigravity
Explore how Gemini 4 Argon in Antigravity autonomously built a 3D Japanese pagoda in Blender, leveraging Nano Banana and multi-agent looping.
Announcing Gemini 4 Carbon, our strongest model with state-of-the-art cybersecurity and long horizon capabilities.
Carbon represents a step up across most benchmarks, delivering strong frontier performance in autonomous coding and cyber defense.
We’re first making Carbon available to a subset of trusted cyber defenders in our Fairwind program. We’ll expand access as soon as possible.
George Kenwright retweeted
Replying to @GeokenAI
@GeokenAI's Argon Pagoda but make it @Aprilzz423’s colorfigures SVG skill with Gemini 4 Argon.
Single shot, simple prompt. Skill + Reference video.
Argon's ability to understand references and recreate with a simple loop are mind blowing. The consistency every time gets me.
So cool!
Introducing tiny.london
A massive isometric pixel map of London, based on amazing NYC demo by @_coenen from 9 months ago
My agent created the dataset, finetuned qwen image LoRA adapter, did a multiday inference run and built an efficient tile viewer!
5
613
George Kenwright retweeted
11 hours with Gemini 4 Argon, Google literally cooked with Gemini Argon. 🔥
it's insane for building 3D stuff. A full Asuka-period Japanese pagoda temple, built in Blender and rendered in Cycles. The level of precision in everything is just crazy.
See a fully-explorable 3D world build itself in minutes ⛩️
Watch @GoogleDeepMind's @GeokenAI explain how Gemini 4 Argon acts as a virtual builder in @antigravity to autonomously create a stunning, multi-tiered pagoda sanctuary.
George Kenwright retweeted
终于完成了,复刻了@GoogleDeepMind发的@GeokenAI 利用Gemini 4 Argon制作的演示视频。昨天看完视频之后非常激动,立马订阅了ultra然后上手尝试,选择了中国传统建筑作为主题。根据Argon自己的统计,花费大约12小时,累计token吞吐量大约100M。
George Kenwright retweeted
the part of argon i love most is the critique loop. one agent literally looks at the render and goes "too blocky" before it gets fixed
so that's the turning point of this video: wooden toy pagoda → photoreal temple, same shot
congrats to the gemini team 👏 made in hyperframes studio with one prompt


























